Reveal the top 10 major trends in the AI industry! Li Feifei's team released the "2024 Artificial Intelligence Index Report"
Ten years ago, the world's most advanced AI systems were unable to match human-level image classification capabilities. They are unable to understand language, have difficulty reasoning visually, and fail even the most basic reading comprehension tests. Today, AI systems routinely outperform humans in standard benchmarks.
Recently, the Stanford University-Human-Centered Artificial Intelligence Institute (Stanford HAI), co-led by Li Feifei's team, recently released the latest Artificial Intelligence Index Report 2024
"The 2024 Index is our most comprehensive report to date, and it comes at a significant time when the impact of AI on society has never been more apparent," Stanford HAI officially said. "The report provides a detailed analysis of the state of artificial intelligence (AI) development in 2023, covering technological advances, public perception, policy dynamics, and the impact of AI on various socio-economic aspects. The report shows that in 2023, AI has made impressive technological breakthroughs.
The latest AI systems, such as GPT-4, Gemini and Claude, demonstrate powerful multimodal capabilities to not only generate fluent multilingual text, but also process images and audio. In some benchmarks, AI has surpassed human levels. However, the report also points out that AI still lags behind humans in more complex tasks, such as high-level math operations, visual common sense reasoning, and planning. Researchers are developing new, more challenging benchmarks to advance AI. In terms of R&D, industry companies will continue to dominate cutting-edge AI research in 2023, producing 51 major machine learning models, compared to only 15 in academia. The report notes that the cost of training top models like GPT-4 and Gemini Ultra also continues to climb.
Number of Notable Machine Learning Models by Industry, 2003-23
New estimates suggest that some cutting-edge systems, such as OpenAI's GPT-4, can cost up to $78 million to train. Google Gemini has a list price tag of $191 million. In comparison, some of the most advanced models released five years ago, the original Transformer model (2017) and RoBERTa Large (2019), cost about $900 and $160,000 to train, respectively.
Estimated training costs for specific AI models in 2017-23
However, there has also been an explosion of open-source AI research. In 2023, the number of AI-related projects on GitHub increased by 59.3%, and the total number of project stars more than tripled.
In the economic and policy realm, generative AI investment soared to $25.2 billion in 2023. Several studies have also shown that AI can improve worker productivity and work quality. At the same time, the United States passed a record number of AI-related regulations last year. However, the report also reveals the public's ambivalence about AI. According to a global survey, 66% of people believe that AI will greatly impact their lives in the next 3 to 5 years, while 52% are anxious about AI products and services, up 13 percentage points from last year. Finally, from a global perspective, the United States continues to lead the way in the field of artificial intelligence. In 2023, significant AI models from U.S. institutions (61) significantly surpassed those of the European Union (21) and China (15). The U.S. also remains a top destination for AI investment. In 2023, U.S. private investment in AI totaled $67.2 billion, nearly nine times that of China.
Number of Notable Machine Learning Models by Geographic Region in 2023
The new index shows that China is the biggest competitor of the United States. China has the highest number of robot installations in the world, with more robots installed in 2023 than the rest of the world combined. Similarly, the majority of the world's AI patents (61%) come from China. In addition, the report reveals the top 10 key trends in the AI industry in 2023:
01 AI surpasses humans in some areas, but it still lags behind humans in complex tasksAI has surpassed humans in tasks such as image classification, visual reasoning, and English comprehension, highlighting AI's significant advantages in processing big data and recognizing patterns. However, AI still lags behind humans in more complex tasks, such as competition math, visual reasoning, and planning. This reveals that AI has certain limitations in areas that require deep understanding and creative thinking.
Comparison of some AI index technical performance benchmarks with human performance
02 Industry leads the cutting-edge development of AIIn 2023, the industry has played a leading role in the field of AI, with a total of 51 well-known AI models launched, while only 15 have been launched by academia. In addition, the collaboration between industry and academia has also yielded remarkable results, with a total of 21 important models being launched. This phenomenon shows that industry-university-research cooperation has great potential and value in promoting the progress of AI technology.
Number of Notable Machine Learning Models by Industry, 2003-23
03 The training cost of advanced AI models continues to rise, and the training cost of AI models has reached an all-time high. For example, OpenAI's GPT-4 model is estimated to cost $78 million, while Google's Gemini Ultra model is as high as $191 million. This rising cost may lead to an excessive concentration of resources in a small number of large enterprises, which in turn will affect the diversity and popularity of AI technology.
Estimated training costs for specific AI models in 2017-23
04 The United States temporarily leads, and China continues to break through in 2023, with the number of top AI models originating from the United States reaching 61; 21 in the European Union; China is catching up with the European Union, and there are already 15 of them. This data shows that the United States continues to maintain its leading position in the global AI field, while China is also continuing to make technological breakthroughs and innovations.
Number of Notable Machine Learning Models by Selected Geographic Region in 2023 In 2023, private investment in AI in the U.S. totaled $67.2 billion, nearly nine times that of China.
2010-22 Status of AI Patents Filed by Selected Geographic Regions
But so far, China is still the biggest competitor of the United States, and China has the largest number of robot installations in the world; Similarly, the majority of the world's AI patents (61%) come from China.
05 Lack of Robust, Standardized Assessment of Responsibility for Large Language Models The latest research on the AI Index shows that there is a serious lack of standardization in responsible AI reporting. Leading developers, including OpenAI, Google, and Anthropic, primarily test their models against different responsible AI benchmarks. This practice complicates the work of systematically comparing the risks and limitations of top-tier AI models.
Responsible AI Benchmark Report for Popular Foundation Models
06 Significant growth in investment in generative AIDespite the overall decline in private investment in AI last year, investment in generative AI has risen to $25.2 billion. In the generative AI space, OpenAI, Anthropic, Hugging Face, and Inflection have all successfully completed large-scale fundraising activities.
Private Investment in Generative AI, 2019-2023
07 AI improves the efficiency and quality of office workers, narrows the skills gapMultiple studies have shown that AI can improve the efficiency and quality of work of office workers, and even close the skills gap between low-skilled and high-skilled workers
。 However, there are also some studies that suggest that if the use of AI is not properly regulated, it can lead to lower performance among office workers.
Cross-study comparison of the speed at which users complete tasks
08 AI Drives Scientific ProgressSince 2022, AI has begun to play an important role in the field of scientific discovery. By 2023, the application of AI in the field of science has become more extensive and deeper, and a series of breakthrough results have been achieved, such as AlphaDev to improve the efficiency of algorithm sequencing and GNoME to promote material discovery.
AlphaDev vs. human benchmark when optimizing algorithm length
09 The number of AI-related regulations continues to growThe number of AI-related regulations in the United States has increased significantly in the past year and the past five years, and China has also introduced a series of AI-related laws and regulations. This trend reflects the importance and actions of governments around the world to regulate the development of AI and protect the public interest.
Number of AI-related regulations in the U.S. from 2016 to 2023
10 Global public awareness and anxiety about the impact of AI are on the rise
According to a survey by market research firm Ipsos, the percentage of people who believe that AI will greatly impact their lives in the next 3-5 years has risen from 60% to 66% in the past year. At the same time, the proportion of people who expressed concern about AI products and services rose by 13 percentage points to 52%. In the U.S., Pew's data shows that the number of people who are worried about AI outnumber those who are excited, rising from 38% in 2022 to 52%. These data show that the global public has a deeper understanding of the potential impact of AI, but it is also accompanied by concern and unease.
Global Perception of Products and Services Using AI, in % of Total, 2022 vs. 2023
Overall, the AI Index 2024 report clearly shows that AI technology is undergoing a seismic transformation, with business leaders, decision-makers and the public around the world beginning to realize its far-reaching implications. In the torrent of new technologies, AI is emerging, and its unlimited potential and broad prospects are gradually being revealed. In the face of this technological revolution, we should adopt an open and inclusive attitude, actively embrace the development of AI, and continuously optimize and improve it in practical applications. AI is not a mysterious technology that is out of reach, and it should not be overly limited.
On the contrary, we should embrace it with a broad vision and an open mind, and encourage it to roam freely in all fields and find the best scenarios that are closely integrated with human life. With the rapid development of technology and the continuous expansion of AI application scenarios, we have reason to believe that AI will play an increasingly important role in work and life in the future. It will be our right-hand man in solving problems, a key force in improving productivity, and a valuable partner in optimizing the quality of life. Through continuous technological innovation and practical exploration, we will have a deeper understanding of the value of AI and better use it to serve humanity. Let's work together to welcome the arrival of the AI era with a positive attitude and jointly create a smarter, more efficient, and better future.